def classify(sample_id, from_name='', *args, **kwargs):
"""
Classifies given samples
"""
class_sample = ClassifiedSample.objects.get(id=sample_id)
if class_sample.label:
return
job = class_sample.job
# If classifier is not trained, return - it will be reclassified if
# the classifier finishes training
if not job.is_classifier_trained():
return
classifier = classifier_factory.create_classifier(job.id)
label = classifier.classify(class_sample)
if label is None:
# Something went wrong
log.warning(
'[Classification] Got None label for sample %d. Retrying.' %
class_sample.id)
current.retry(
countdown=min(60 * 2**(current.request.retries % 6), 60 * 60 * 1),
max_retries=None,
)
ClassifiedSample.objects.filter(id=sample_id).update(label=label)
send_event(
'EventSampleClassified',
job_id=job.id,
class_id=class_sample.id,
sample_id=class_sample.sample.id,
)
def setUp(self):
self.u = User.objects.create_user(username='******', password='******')
self.job = Job.objects.create_active(
account=self.u.get_profile(),
gold_samples=[{'url': '10clouds.com', 'label': LABEL_YES}])
self.train_data = [
Sample(job=self.job, source_type='',
text='Mechanical squirrel screwdriver over car'),
Sample(job=self.job, source_type='',
text='Screwdriver fix mechanical bike bolts'),
Sample(job=self.job, source_type='',
text='Brown banana apple pinapple potato'),
Sample(job=self.job, source_type='',
text='apple pinapple potato'),
Sample(job=self.job, source_type='',
text='Hippo tree over lagoon'),
Sample(job=self.job, source_type='',
text='Green tan with true fox')
]
self.labels = [LABEL_YES, LABEL_YES, LABEL_NO, LABEL_NO, LABEL_NO, LABEL_NO]
self.classified = []
for idx, sample in enumerate(self.train_data):
self.classified.append(ClassifiedSample.objects.create(
job=self.job,
sample=sample,
label=self.labels[idx]
))
self.classifier247 = classifier_factory.create_classifier(
job_id=self.job.id,
)
def classify(sample_id, from_name='', *args, **kwargs):
"""
Classifies given samples
"""
class_sample = ClassifiedSample.objects.get(id=sample_id)
if class_sample.label:
return
job = class_sample.job
# If classifier is not trained, return - it will be reclassified if
# the classifier finishes training
if not job.is_classifier_trained():
return
classifier = classifier_factory.create_classifier(job.id)
label = classifier.classify(class_sample)
if label is None:
# Something went wrong
log.warning(
'[Classification] Got None label for sample %d. Retrying.' % class_sample.id
)
current.retry(
countdown=min(60 * 2 ** (current.request.retries % 6), 60 * 60 * 1),
max_retries=None,
)
ClassifiedSample.objects.filter(id=sample_id).update(label=label)
send_event(
'EventSampleClassified',
job_id=job.id,
class_id=class_sample.id,
sample_id=class_sample.sample.id,
)
def testMetrics(self):
# Mock classifier.analyze() method so we don't use up resources
classifier = classifier_factory.create_classifier(self.job.id)
new_analyze = lambda: {
'modelDescription': {
'confusionMatrix': {
LABEL_YES: {
LABEL_YES: 5.0,
LABEL_NO: 3.0,
},
LABEL_NO: {
LABEL_YES: 2.0,
LABEL_NO: 7.0,
}
}
}
}
classifier.analyze = new_analyze
update_classifier_stats(classifier, self.job)
# 1 is from SimpleClassifier train on create,
# 1 is from update_classifier_stats above
self.assertEqual(ClassifierPerformance.objects.count(), 2)
cp = ClassifierPerformance.objects.filter(job=self.job).order_by('-id')
cp = cp[0]
metrics_to_check = (
'TPR',
'TNR',
'AUC',
)
for metric in metrics_to_check:
self.assertIn(metric, cp.value)
def testMetrics(self):
# Mock classifier.analyze() method so we don't use up resources
classifier = classifier_factory.create_classifier(self.job.id)
new_analyze = lambda: {
'modelDescription': {
'confusionMatrix': {
LABEL_YES: {
LABEL_YES: 5.0,
LABEL_NO: 3.0,
},
LABEL_NO: {
LABEL_YES: 2.0,
LABEL_NO: 7.0,
}
}
}
}
classifier.analyze = new_analyze
update_classifier_stats(classifier, self.job)
# 1 is from SimpleClassifier train on create,
# 1 is from update_classifier_stats above
self.assertEqual(ClassifierPerformance.objects.count(), 2)
cp = ClassifierPerformance.objects.filter(job=self.job).order_by('-id')
cp = cp[0]
metrics_to_check = (
'TPR',
'TNR',
'AUC',
)
for metric in metrics_to_check:
self.assertIn(metric, cp.value)
def train(set_id):
training_set = TrainingSet.objects.get(id=set_id)
job = training_set.job
classifier = classifier_factory.create_classifier(job.id)
samples = (training_sample
for training_sample in training_set.training_samples.all())
classifier.train(samples, set_id=set_id)
job = Job.objects.get(id=job.id)
if job.is_classifier_trained():
send_event(
"EventClassifierTrained",
job_id=job.id,
)
def testClassifierFactory(self):
job = Job.objects.create_active(
account=self.u.get_profile(),
gold_samples=[{'url': '10clouds.com', 'label': LABEL_YES}],
)
factory = classifier_factory.create_classifier(job.id)
self.assertEqual(factory.__class__, Classifier247)
self.assertEqual(Classifier.objects.filter(
job=job,
type='SimpleClassifier',
).count(), 2)
cs = ClassifiedSample.objects.create_by_owner(
job=job,
url='http://google.com',
)
cs = ClassifiedSample.objects.get(id=cs.id)
self.assertTrue(factory.classify(cs))
def train(set_id):
training_set = TrainingSet.objects.get(id=set_id)
job = training_set.job
classifier = classifier_factory.create_classifier(job.id)
samples = (training_sample
for training_sample in training_set.training_samples.all())
classifier.train(samples, set_id=set_id)
job = Job.objects.get(id=job.id)
if job.is_classifier_trained():
send_event(
"EventClassifierTrained",
job_id=job.id,
)
def classify_btm(sample_id, from_name='', *args, **kwargs):
"""
Classifies given samples
"""
log.info(
'[BTMClassification] Got sample %d for classification.' % sample_id
)
btm_sample = BeatTheMachineSample.objects.get(id=sample_id)
if btm_sample.label:
return
job = btm_sample.job
# If classifier is not trained, retry later
if not job.is_classifier_trained():
current.retry(countdown=min(60 * 2 ** current.request.retries,
60 * 60 * 24))
classifier = classifier_factory.create_classifier(job.id)
label = classifier.classify(btm_sample)
if label is None:
# Something went wrong
log.warning(
'[BTMClassification] Got None label for sample %d. Retrying.'
% btm_sample.id
)
current.retry(countdown=min(60 * 2 ** current.request.retries,
60 * 60 * 24))
BeatTheMachineSample.objects.filter(id=sample_id).update(label=label)
btm_sample.updateBTMStatus()
send_event(
'EventSampleBTM',
job_id=job.id,
btm_id=btm_sample.id,
sample_id=btm_sample.sample.id,
)
def classify_btm(sample_id, from_name='', *args, **kwargs):
"""
Classifies given samples
"""
log.info('[BTMClassification] Got sample %d for classification.' %
sample_id)
btm_sample = BeatTheMachineSample.objects.get(id=sample_id)
if btm_sample.label:
return
job = btm_sample.job
# If classifier is not trained, retry later
if not job.is_classifier_trained():
current.retry(countdown=min(60 * 2**current.request.retries, 60 * 60 *
24))
classifier = classifier_factory.create_classifier(job.id)
label = classifier.classify(btm_sample)
if label is None:
# Something went wrong
log.warning(
'[BTMClassification] Got None label for sample %d. Retrying.' %
btm_sample.id)
current.retry(countdown=min(60 * 2**current.request.retries, 60 * 60 *
24))
BeatTheMachineSample.objects.filter(id=sample_id).update(label=label)
btm_sample.updateBTMStatus()
send_event(
'EventSampleBTM',
job_id=job.id,
btm_id=btm_sample.id,
sample_id=btm_sample.sample.id,
)
def testGoogleP(self):
results = {
'insert': '',
'analyze': {
'modelDescription': {
'confusionMatrix': {
LABEL_YES: {
LABEL_YES: 1,
LABEL_NO: 0,
},
LABEL_NO: {
LABEL_YES: 0,
LABEL_NO: 1,
}
}
}
},
}
class MockGooglePrediction(object):
def trainedmodels(self, *args, **kwargs):
return self
def insert(self, *args, **kwargs):
self.method = 'insert'
return self
def analyze(self, *args, **kwargs):
self.method = 'analyze'
return self
def execute(self, *args, **kwargs):
result = results[self.method]
if isinstance(result, Exception):
raise result
return result
def predict(self, *args, **kwargs):
self.method = 'predict'
return self
def get(self, *args, **kwargs):
self.method = 'get'
return self
def build(*args, **kwargs):
return MockGooglePrediction()
target = 'urlannotator.main.factories.settings.JOB_DEFAULT_CLASSIFIER'
self.patch = mock.patch(target, new='GooglePredictionClassifier')
self.patch.start()
target = 'urlannotator.classification.classifiers.build'
self.patch_api = mock.patch(target, new=build)
self.patch_api.start()
target = 'urlannotator.classification.classifiers.GSConnection'
self.patch_bucket = mock.patch(target)
self.patch_bucket.start()
target = 'urlannotator.classification.classifiers.Key'
self.patch_key = mock.patch(target)
self.patch_key.start()
u = User.objects.create_user(username='******', password='******')
job = Job.objects.create_active(
account=u.get_profile(),
gold_samples=[{'url': '10clouds.com', 'label': LABEL_YES}])
classifier = classifier_factory.create_classifier(job.id)
classifier.analyze()
results['analyze'] = Exception()
classifier.analyze()
results['get'] = {'trainingStatus': 'test'}
self.assertEqual(classifier.get_train_status(), 'test')
results['get'] = Exception()
self.assertEqual(classifier.get_train_status(), CLASS_TRAIN_STATUS_RUNNING)
results['get'] = {'trainingStatus': 'ERROR: test'}
with self.assertRaises(ClassifierTrainingCriticalError):
classifier.get_train_status()
results['get'] = Exception()
train_set = job.trainingset_set.all()[0]
classifier.train(samples=train_set.training_samples.all())
classifier.train(set_id=train_set.id, turn_off=True)
job.set_classifier_trained()
results['predict'] = {'outputLabel': LABEL_YES, 'outputMulti': [{'label': LABEL_YES, 'score': 1},{'label': LABEL_NO, 'score': 0}]}
cs = ClassifiedSample.objects.create_by_owner(
job=job,
url='http://google.com',
)
# Refresh the Classified Sample
cs = ClassifiedSample.objects.get(id=cs.id)
self.assertEqual(classifier.classify(sample=cs), LABEL_YES)
self.assertEqual(classifier.classify_with_info(sample=cs), results['predict'])
# What if we remove the classfier's id?!?!
classifier.model = None
self.assertEqual(classifier.classify(sample=cs), None)
self.assertEqual(classifier.classify_with_info(sample=cs), None)